INTERVIEWS · Written and reviewed by the Media2Text team · Published November 3, 2026 · 9 minute read

How to transcribe an interview accurately

A good interview transcript lets you quote people accurately and find evidence quickly. This guide walks through the whole process, from recording to a checked transcript you can cite, for researchers, journalists, students and anyone who interviews people for a living.

Automatic speech recognition now produces a strong first draft in minutes. The work that remains is the part software cannot do for you: deciding what kind of transcript you need, confirming who said what and checking the passages you will actually use. Each section below covers one of those decisions.

Before the interview: record for the transcript

Most transcription problems start at the recording. A few minutes of preparation saves far more time in review.

Put the microphone close to the speakers

Distance matters more than equipment. A phone placed on the table between two people usually beats an expensive microphone across the room. For a remote interview, ask the guest to use headphones so their voice is not picked up twice, and record locally if your call software allows it. A separate track for each participant makes speaker separation much easier.

Reduce background noise you can control

Choose a quiet room, close windows and avoid spaces with hard walls that echo. Cafés are pleasant for conversation and difficult for speech recognition: espresso machines, music and nearby voices all compete with your interviewee. If you must record in public, sit away from speakers and kitchens.

Say names and key terms clearly at the start

Begin with a short introduction: the date, the interviewee's name and how it is spelled, their role and any organisation names you expect to come up. This gives you a reference point in the audio and makes later corrections faster. It also confirms consent on the record if your project requires it.

Keep the original file

Save the original recording before you trim or convert it. If you later need to check a disputed quote, the untouched file is your best evidence. Upload the original rather than a re-compressed copy; converting MP3 to WAV does not restore lost detail.

Choose a transcript style before you start

There is no single correct transcript. The right style depends on how the text will be used, and choosing early keeps your editing consistent.

Speech recognition output is usually closest to clean verbatim: it captures the words but often drops many fillers. If your method requires full verbatim, plan extra listening time to add hesitations and overlaps back in. Whatever you choose, write the rule down so that everyone working on a project transcribes the same way.

Transcribing the interview step by step

1. Upload the recording. Media2Text accepts common audio formats such as MP3, M4A, WAV, AAC and FLAC, and video formats such as MP4 and MOV. For video, the audio track is extracted in your browser and the original video stays on your device. Files can be up to 6 GB.
2. Set the language. Automatic language identification works well for most interviews. Set the language manually if the recording opens with silence, music or a short greeting in another language. Media2Text documents 60 transcription languages, and interviews that switch between two languages can still be transcribed, though code-switched passages deserve extra review.
3. Keep speaker labels on. Automatic speaker labels split the transcript into turns such as Speaker 1 and Speaker 2. They describe different voices, not real identities, so you will name them yourself in review.
4. Wait for processing. Results are usually ready within minutes, depending on length and current load. A typical one-hour interview does not need to be watched while it processes.

If your interview is already published online as a public video, you may be able to paste the link instead of uploading a file. The video link to text page lists the supported platforms and the limits on private or restricted media.

Review: where the real accuracy comes from

Review is where a draft becomes a transcript you can rely on. Work in passes rather than trying to fix everything at once.

Pass 1: name the speakers

Listen to the first few turns and rename the generic labels to the interviewer and the interviewee, or to pseudonyms if your research requires anonymity. Check every place where voices overlap or where one person speaks very briefly, because these are where automatic labels most often merge or split.

Pass 2: fix repeated terms with find and replace

Names of people, places, products and organisations are the most common recognition errors, and they usually repeat. Use find and replace to correct a misspelled name everywhere at once, after confirming that every match refers to the same thing.

Pass 3: check the passages you will quote

You do not need to listen to an entire hour again. Mark the passages you plan to quote, cite or code, then click each timestamp and listen. Pay close attention to numbers, dates, negations such as "did" versus "didn't", and words that sound alike. A quote that reads plausibly can still be wrong.

Pass 4: spot-check the rest

Listen to the beginning, a section in the middle and the end even if the transcript reads well. This catches a change in recording conditions, a stretch of crosstalk or missing final words.

Time budget: for a clear two-person recording, targeted review often takes a fraction of the interview length. Noisy audio, strong overlap or specialist vocabulary can make review take as long as the recording itself. Plan accordingly before a deadline.

Export for analysis, writing or archiving

Choose the export based on what happens next:

• DOCX or PDF for sharing with co-authors, supervisors or editors, with speaker names and timestamps included.
• TXT or Markdown for importing into writing tools, note apps or qualitative analysis software that accepts plain text.
• CSV, TSV or JSON when you want each segment, speaker and timestamp as structured data for a spreadsheet or your own scripts.
• SRT or VTT if the interview will be published as a video with captions.

For long projects, keep a consistent file name that links each export to the original recording, the interview date and the participant code. With an account you can also generate an Interview-style AI summary with source references, which is useful for a first overview but never a substitute for checking quotes in the transcript. The summary, mind map and translation guide explains those tools.

Interviews in more than one language

If you interview someone in a language your readers or co-researchers do not speak, transcribe in the original language first, then create a translated track. The translation keeps the original speaker labels and timestamps, so you can view both versions side by side and return to the exact moment in the audio. Quote from the original wherever possible and treat the translation as a working aid, especially for idioms and culturally specific terms.

For Filipino and Taglish interviews, see the Tagalog audio to text guide, which covers code-switching in more detail.

Consent, privacy and data handling

Interview recordings often contain personal information. Before you record, make sure participants know how the recording and transcript will be used and stored, and follow your institution's ethics approval or your newsroom's standards. When you upload, choose a service whose data handling you understand: Media2Text extracts audio from supported videos in the browser, and the prepared audio is retained according to the published privacy policy. The security and data flow page explains what leaves your device.

After the project, delete recordings and transcripts you no longer need, and remove identifying details from any excerpts you publish if participants were promised anonymity.

Interview transcription checklist

1. Record close to the speakers in a quiet room, and keep the original file.
2. State names, roles and key terms at the start of the recording.
3. Decide on full verbatim, clean verbatim or edited style.
4. Upload the original recording and confirm the language.
5. Rename speakers and check overlaps.
6. Correct repeated names with find and replace.
7. Listen to every passage you will quote.
8. Export DOCX for sharing, CSV or JSON for analysis, SRT for captions.
9. Store and delete files according to your consent agreement.

Common questions

How long does it take to transcribe a one-hour interview?

Typing a one-hour interview by hand often takes several hours. With automatic transcription, the draft is usually ready within minutes, and the remaining time is review. How long review takes depends mainly on audio quality and how many passages you need to verify.

Can automatic transcription tell who is speaking?

It can separate different voices into labelled turns, but it does not know who the people are. You assign names during review. Similar voices, very short turns and crosstalk can cause labels to merge, so check them.

Is an automatic transcript accurate enough to quote?

Use it as a draft. Always check a quote against the audio before publishing or citing it, especially names, numbers and negations.

Can I try it before paying?

Yes. The free trial covers up to 2 lifetime transcription tasks sharing 75 minutes in total, with no credit card. Paid plans add unlimited uploads and files up to 5 hours 15 minutes long.

Start with your next interview

Upload a recording on the audio to text page, or use video to text for recorded video calls. For multi-speaker sessions such as focus groups, the speech to text page covers speaker review in more detail.